A Screening Method for Determining Left Ventricular Systolic Function Based on Spectral Analysis of a Single-Channel...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Diagnosing heart failure is complex and symptoms are often non-specific, limiting screening applicability.
• A simple, non-invasive screening method for systolic heart dysfunction using readily available biosignals is needed.
• Single-channel electrocardiogram (ECG) analysis offers a potential solution for accessible cardiac screening.

Purpose of the Study:

• To develop a machine learning-based screening model for left ventricular systolic dysfunction.
• To analyze single-channel ECG parameters for detecting reduced left ventricular ejection fraction (LVEF).
• To create a screening tool that does not require medical staff involvement.

Main Methods:

• Inclusion of 624 patients (18-90 years) undergoing echocardiography and single-channel I-lead ECG.
• Determination of left ventricle ejection fraction (LV EF) using the BIPLANE Simpson method.
• Advanced signal processing and machine learning algorithms (Lasso regression, Extra Trees) applied to ECG data.

Main Results:

• Lasso regression achieved 79.2% sensitivity and 81.7% specificity for LVEF <52% (men) / <54% (women) (AUC=0.849).
• Extra Trees model showed 83.1% sensitivity and 82.7% specificity for LVEF <40% (AUC=0.972).
• External validation on 600 patients demonstrated 98% accuracy, 98.4% specificity, and 93.5% sensitivity.

Conclusions:

• Machine learning analysis of single-channel ECG parameters offers high diagnostic accuracy for screening left ventricular systolic dysfunction.
• The developed model shows potential for a simple, effective, and non-invasive method for early detection of cardiac issues.
• Modern signal processing and AI technologies can significantly enhance cardiovascular screening capabilities.